Smaller AI models can now learn specific skills without forgetting others

Current methods for shrinking large AI models often make them forget some skills or waste training effort. This paper introduces a new way to shrink AI models that helps them keep their specific skills and uses training resources more efficiently.

Making powerful AI models smaller and cheaper to run is a major goal for developers. Until now, shrinking a large AI model often meant sacrificing some of its abilities or spending a lot of time trying to balance different skills. This new method offers a more intelligent way to compress models, allowing developers to create specialized AI that is both powerful and efficient.